Snorkel AI raised a $350 million Series E at a $3.5 billion valuation, the seven-year-old startup announced, roughly tripling the $1.3 billion price tag it carried when it closed a $100 million Series D just 17 months ago. Insight Partners and S32 led the round, with existing backers Addition, Lightspeed, Greylock, GV, and Wells Fargo participating. The company said its annualized revenue run-rate now stands at $375 million, an 18-fold jump over the last 12 months.
That growth curve is what's driving the valuation. Snorkel sells training data and reinforcement learning environments to AI labs and large enterprises building their own models — a market that barely existed at commercial scale two years ago and now underwrites a cluster of billion-dollar businesses. The pitch to investors is straightforward: every frontier lab needs more high-quality data than it can produce internally, and it needs it in domain-specific shapes that generic web scrapes cannot deliver.
Snorkel started life in a different business. Founded by CEO Alex Ratner following four years of research at a Stanford AI lab, the company launched commercially in 2019 selling software that automated data labeling — customers ran the tooling in-house and produced their own datasets. Last year Snorkel flipped the model, and now delivers finished datasets and RL environments directly, an offering it calls data-as-a-service.
Key facts
- 01Snorkel AI raised a $350M Series E at a $3.5B valuation, nearly triple its $1.3B mark from 17 months ago.
- 02Revenue run-rate hit $375M, an 18x increase over the last 12 months.
- 03Insight Partners and S32 led the round; Addition, Lightspeed, Greylock, GV, and Wells Fargo also participated.
- 04Rivals Mercor ($2B), Handshake ($1B), and Micro1 ($500M) have seen similar surges in gross annualized revenue.
- 05Snorkel pivoted last year from data-labeling software to a data-as-a-service model selling completed datasets and RL environments.
The hybrid production method is the differentiator Snorkel is leaning on. Rather than operating as a pure human-expert marketplace, the company combines its own software and models to generate data synthetically alongside subject-matter experts. That mix, Snorkel argues, gives it better unit economics than competitors that route the majority of contract value straight through to human labelers.
The unit-economics point matters when comparing headline numbers. Mercor's gross annualized revenue has climbed to $2 billion, Handshake crossed $1 billion earlier this year, and Micro1 has scaled to $500 million. But those firms pay roughly 60% to 70% of their top-line revenue directly to the domain specialists doing the work, meaning net revenue is substantially lower than the reported gross figures.
Snorkel, by contrast, books human-expert payments as cost of goods sold rather than netting them against revenue, since it is selling completed datasets and RL environments rather than reselling human labor. The company said that accounting treatment reflects the structure of the product — customers buy the artifact, not the hours. That framing lets Snorkel post a $375 million run-rate that is more directly comparable to a software business than to a services shop.
The demand backdrop explains why five separate companies are scaling this fast at once. Frontier labs have moved past the era when a bigger web crawl solved the next capability jump; performance gains now come from post-training on carefully specified data — expert reasoning traces, coding trajectories, RL environments where a model can be graded on outcomes rather than tokens. That work cannot be crowdsourced from generalists, and labs are increasingly unwilling to build the operations in-house.
Insight and S32 leading the round is a signal in itself. Insight has historically been a late-stage software investor that underwrites growth against durable ARR, not a typical AI-round participant chasing model bets. The firm's willingness to lead at $3.5 billion suggests it views Snorkel's revenue as durable enterprise contract value rather than volatile lab spend that could evaporate if training budgets contract.
The risk on the other side is exactly that concentration. AI-lab training budgets are enormous but concentrated across a handful of buyers, and any single lab pulling data work back in-house — or shifting to a competitor with a better price on the same specialist pool — would show up quickly in run-rate. The multi-vendor pattern across Snorkel, Mercor, Handshake, and Micro1 hints that labs are deliberately spreading spend, which cuts both ways for any one supplier.
The data-labeling category is now the highest-margin infrastructure layer in the AI stack that isn't chips. Snorkel's bet — that owning the software and synthetic-data pipeline lets it capture more of the value than pure marketplaces can — is the one to watch. If the $375 million run-rate holds its growth rate into next year, the $3.5 billion price will look conservative; if labs start insourcing, this cohort compresses fast, and Snorkel's software-heavy model becomes the only defensible one in the group.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
